---
title: How do you govern Snowflake Intelligence agents?
url: https://opencontroller.lyzr.com/blog/govern-snowflake-intelligence
published: 2026-11-17
updated: 2026-11-17
author: Lyzr
---

# How do you govern Snowflake Intelligence agents?

**Snowflake Intelligence agents run with the end user's own credentials, so governance comes down to four grants: who holds the Cortex database role, who has USAGE on each agent, what the user's role can read through each tool, and which agents are listed on the account's Snowflake Intelligence object. Add a Restricted Session Scope to cap what any agent session can do, even when the user's role could do more.**

Snowflake Intelligence is Snowflake's interface for chatting with data agents built on Cortex Agents. Snowflake's current documentation labels the interface Snowflake CoWork, while SQL objects keep Snowflake Intelligence names such as `SNOWFLAKE_INTELLIGENCE_OBJECT_DEFAULT` and the `CREATE SNOWFLAKE INTELLIGENCE` privilege. Whichever name your teams use, the controls below are the same.

## Who can use an agent, and what can it reach?

Access is layered. Each layer is a normal Snowflake grant.

| Layer | Grant | Effect |
|---|---|---|
| Cortex features | `SNOWFLAKE.CORTEX_USER` database role (granted to PUBLIC by default) or the narrower `SNOWFLAKE.CORTEX_AGENT_USER` | Whether the role can use Cortex agents at all |
| Agent visibility | USAGE on the database and schema | Whether the user can see the agent object |
| Agent use | USAGE on the agent | Whether the user can query it |
| Tool data | USAGE on Cortex Search services, SELECT on tables, READ on stages, USAGE on semantic views and on custom tools such as stored procedures | What each tool can return or run for that user |
| Listing | USAGE on the Snowflake Intelligence object, plus the agent being added to it | Whether the agent appears in the agent list |

The most important line in Snowflake's documentation is that all queries from the interface use the user's credentials. Role-based access control and masking policies for that user apply. An agent cannot read a table the user's role cannot read. It can, however, read everything the user's role can, which is why role hygiene matters more once agents arrive. See [how masking policies apply to Cortex Agents](/blog/snowflake-masking-policies-and-cortex-agents).

## What should you change from the defaults?

1. **Revoke `CORTEX_USER` from PUBLIC** if not everyone should use Cortex AI, and grant it, or `CORTEX_AGENT_USER`, to specific roles. `CORTEX_AGENT_USER` gives access to the Cortex Agents API without the other Cortex features.
2. **Create the Snowflake Intelligence object** so only agents you add are listed. Without it, agents are reachable by direct link or in Snowsight. Only one object per account is allowed, and creating it needs `CREATE SNOWFLAKE INTELLIGENCE ON ACCOUNT`, which ACCOUNTADMIN holds by default.
3. **Add agents deliberately** with `ALTER SNOWFLAKE INTELLIGENCE SNOWFLAKE_INTELLIGENCE_OBJECT_DEFAULT ADD AGENT <db.schema.agent>`. The person adding needs modify rights on the object and USAGE on the agent.
4. **Set default roles and warehouses.** The interface starts each session with the user's default role and warehouse. A default role that is broader than the user's day-to-day role widens what agents can read.
5. **Grant tool objects to roles, not users,** and keep write-capable stored procedures in separate roles from read-only tools.

## How do Restricted Session Scopes limit agent sessions?

Snowflake has announced general availability of Restricted Session Scopes (RSS), a privilege ceiling for agent sessions. Per Snowflake, an RSS can only restrict what an agent does on a user's behalf, never expand it. It applies to sessions Snowflake identifies as agentic, including Cortex Agents, Snowflake CoWork, MCP integrations, and external agents that identify as agentic through OAuth or the `SERVICE_AGENT` user type.

An admin creates a `RESTRICTED SESSION SCOPE` object listing allowed privileges and blocked roles, then attaches it through a session policy to the account or a user. Typical limits:

- Block writes, or allow writes only to a named database
- Block administrative roles such as ACCOUNTADMIN, SYSADMIN and SECURITYADMIN
- Block data-governance roles that unmask PII
- Disallow role switching

The built-in `SNOWFLAKE$DATA_READ_WITH_AI` scope enforces read-only access while keeping AI function use. For most analytics agents, that is a sensible starting ceiling.

## How do you monitor usage and cost?

Agent logs show prompts, responses and errors, and user feedback on responses is available through Snowflake's Cortex Agent monitoring. Review them for unanswered questions and for tools returning errors, which often signal a missing grant. For spend, use the views covered in [tracking Snowflake Cortex AI spend](/blog/track-snowflake-cortex-ai-spend), and keep [grants for usage monitoring](/blog/snowflake-grants-for-ai-usage-monitoring) with a small admin group.

## How Opencontroller does this

Opencontroller keeps a governed registry of agents, versions, tools and MCP servers across platforms, flags risky permission and configuration settings through Posture, and applies policy as code with built-in mappings to OWASP, MITRE ATLAS, NIST and ISO 42001. LLM judges score production traffic and live guardrails check groundedness and task adherence. See [how to govern AI agents across platforms](/blog/how-to-govern-ai-agents-across-platforms), the [AI agent control plane overview](/blog/what-is-an-ai-agent-control-plane), or [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Whose permissions does a Snowflake Intelligence agent use?**
The end user's. Snowflake states that all queries use the user's credentials, so that user's role-based access control and masking policies apply.

**How do I stop every user from seeing every agent?**
Create the account's Snowflake Intelligence object and add only approved agents to it. Users then need USAGE on the object to see the list and USAGE on each agent to use it.

**Can I make agent sessions read-only in Snowflake?**
Yes. A Restricted Session Scope such as the built-in `SNOWFLAKE$DATA_READ_WITH_AI` enforces read-only access for sessions identified as agentic.

## Key facts

Snowflake Intelligence (interface labelled Snowflake CoWork in current docs; SQL objects keep SNOWFLAKE_INTELLIGENCE names): queries use the user's credentials so RBAC and masking apply. Access: SNOWFLAKE.CORTEX_USER (granted to PUBLIC by default) or SNOWFLAKE.CORTEX_AGENT_USER; USAGE on database/schema and agent; tool privileges (Cortex Search USAGE, table SELECT, stage READ, semantic view USAGE). One Snowflake Intelligence object per account (SNOWFLAKE_INTELLIGENCE_OBJECT_DEFAULT), needs CREATE SNOWFLAKE INTELLIGENCE ON ACCOUNT; ALTER ... ADD AGENT. Sessions start with default role and warehouse. Restricted Session Scopes (GA per Snowflake blog): privilege ceiling for agentic sessions (Cortex Agents, CoWork, MCP, external agents via OAuth or SERVICE_AGENT), attached by session policy; built-in SNOWFLAKE$DATA_READ_WITH_AI read-only. Opencontroller: registry, Posture, policy as code, LLM judges, guardrails. Pricing starts from $250K per deployment per year.

## Related

- [How do you govern Snowflake Cortex Agents?](https://opencontroller.lyzr.com/blog/govern-snowflake-cortex-agents.md)
- [Do Snowflake masking and row access policies apply to Cortex Agents?](https://opencontroller.lyzr.com/blog/snowflake-masking-policies-and-cortex-agents.md)
- [How do you track Snowflake Cortex AI spend by agent or team?](https://opencontroller.lyzr.com/blog/track-snowflake-cortex-ai-spend.md)
- [More on Platform Guides](https://opencontroller.lyzr.com/blog/cluster/platform-guides)
